Physicochemical Properties and Structure of FLiBeTh Salts: Insights from Machine Learning Accelerated Molecular Dynamics Simulations.

Abstract

LiF-BeF₂-ThF₄ (FLiBeTh) is a promising fuel salt for thorium-based molten salt reactors due to its excellent neutron economy and adjustable properties. However, experiments on such systems remain challenging due to high temperature, corrosiveness, and toxicity. To address these challenges, this study employs molecular dynamics simulations based on a machine learning potential. Using data sets from ab initio calculations and an iterative workflow, a highly accurate machine-learning model was developed, achieving energy and force prediction errors below 1 meV/atom and 50 meV/Å, respectively. This model accurately reproduces the AIMD-predicted radial distribution functions, coordination numbers, and angular distributions. Furthermore, MLMD simulations enabled the exploration of larger-scale or long-term structural characteristics, including coordination shell lifetime, ionic network formation, and physicochemical properties such as density, ionic diffusion, shear viscosity, and thermal conductivity. Results show that increasing ThF₄ concentration promotes the formation of networks composed of Be²⁺, Th⁴⁺, and F^- ions, which significantly reduces ion mobility and changes the physicochemical properties of the molten salts.